I'm trying to help my daughter with the concept of lower and upper bounds of a number when given a specific accuracy, but I'm starting to realise how many holes there are in my own knowledge!
For a specific number of decimal places this is straightforward:
$x=1.23$ to 2 decimal places.
Therefore: $1.225 \le x \lt 1.235$
Equally, for numbers given to a number significant figures it is much the same. Eg:
$x=99$ to 2 significant figures
Therefore: $98.5 \le x \lt 99.5$
Or $x=110$ to 2 significant figures
Therefore: $105 \le x \lt 115$
However, this is what I can't figure out:
$x=100$ to 2 significant figures.
Since anything between 100 and 105 rounded to 2 significant figures would be 100, logically the upper bound is 105.
On the other hand, the lower bound cannot correspondingly be 95 because 95 to 2 significant figures is 95!
Therefore, should it be that in this case:
$99.5 \le x \lt 105$? If so, that rather lacks the symmetry I had expected.
Your analysis is spot on. I suspect most people who talk of significant figures have not thought about this corner case at all. Another asymmetry is that $99$ to two significant figures is $\pm 0.5\%$ but $110$ to two figures is $\pm 5\%$
Most error analysis that I have done or seen has been rather informal. One doesn't worry too much about whether the error bounds are hard limits (like if you measure a length with a ruler marked in mm, the result should be $\pm 0.5$ mm with a uniform distribution) or a standard deviation. If it is a standard deviation, there is no hard limit. Now when you compute with the numbers, we just keep the same number of significant figures if we multiply, use the larger absolute error if we add, and so on. The result is a pretty good of the final error, but it could be outside that. To do a better job, you can use interval arithmetic, keeping track of the maximum and minimum possible. The bounds may well not be symmetric around your calculated value.